Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/7287
標題: 適用於智慧型視訊系統之物體追蹤演算法及晶片實作設計
Object Tracking Algorithm and VLSI Implementation for Advanced Video-based Surveillance System
作者: 鍾育杰
Chung, Yu-Chieh
關鍵字: pan-tilt-zoom camera;視訊;multi-objects detect;wide-angle camera;video frame partition;Object Tracking Algorithm;智慧型;目標物;演算法;攝影機;移動偵測;目標物追蹤
出版社: 電機工程學系
摘要: 
智慧型視訊監控系統一直是近年來相當熱門的研究項目,其目的大都是為了達到目標物之偵測與追蹤為主。視訊監控系統依照環境不同可分成室內及室外兩種,室內監控著重在身份辨識的目的,而室外監控系統其重點大都著重在行為分析上。大部分的視訊監控系統,都只用一台或多台相同的攝影機來取像,在此種情況下有時並無法發揮其最大的功效,本研究利用兩種不同功能的攝影機:廣角攝影機(wide-angle cameras)及二維旋轉變焦式攝影機(pan-tilt-zoom camera)建構室外/內環境監控系統,利用兩種攝影機不同特性的優缺點作互補,以達到行為分析和身份辨識的效果。在本篇論文中,使用實驗室自行研發的影像處理監控平台,包含前端視訊A/D轉換IC和執行智慧型監控演算法的FPGA及周邊電路來建構獨立運作的監視系統。
在前景物偵測部分,利用對廣角攝影機之影像作背景相減來得到目標物位置並持續追蹤,而背景建立的方法則使用漸進式的背景影像建構法,以期能快速建立並更新背景影像。在前景物件的分類本研究提出一個積木式連接元件匹配法.這是應用在電路設計的一種快速低成本,而且是一次完成的物件掃描,不需傳統四連通或八連通的連接物件偵測方式.
在多目標偵測部份,則是利用亮度差異資訊及分割畫面的幾何條件,快速在廣角攝影機之影像中尋找移動人物位置,再利用PTZ攝影機鎖定目標並放大影像。本研究亦提出一個旋轉變焦式攝影機(pan-tilt-zoom camera)穩定物件追蹤的演算法,且相當適用在電路設計上.實驗結果顯示本系統對於以多目標偵測及辨識為目的之室外監控系統有相當的可行性。

This research makes use of two kinds of cameras of different functions: wide-angle camera and pan-tilt-zoom camera to construct the outdoors /indoors surveillance system, The merit and shortcoming that makes use of two different camera characteristics to repair with each other, in order to reach the result of behavior analysis and the body recognizes. In this thesis, we uses the self-develop surveillance platform in our Lab, including the source video A/D signal transform IC, the FPGA implementation of the intelligent surveillance algorithm and peripheral electric circuit to construct the independent operation of surveillance system.
In foreground detection, make use of the current image to subtract with background of the wide-angle camera to get the objects` locations and to track continuously, And the method of the background establishment then make use of the gradually type of the background image construction method, to wish to create and update the background image quickly. The foreground objects classify method use the building-block based connected components match method. This method is fast, only one-pass scan and low-cost for objects classify, it doesn't need four adjacent or eight adjacent traditional connected components rule.
In multi-objects detection, we make use of the information of luminance discrepancy and geometry conditions of the video frame partition, It looks for locations of the moving persons in the image of the wide-angle camera which quickly makes use of the PTZ camera to lockup object and also to enlarge the image. This thesis also proposed a Smooth PTZ Camera Object Tracking Algorithm. The experimental results show that this system has equal feasibility to take having multi-object detect and recognize of the purpose as the outdoors surveillance system.
URI: http://hdl.handle.net/11455/7287
Appears in Collections:電機工程學系所

Show full item record
 

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.